Pedestrian Trajectory Prediction Based on Environmental Shape with Deep Learning

Abstract

In this paper, we propose a trajectory prediction method in an environment where pedestrians are surrounded by obstacles. When an autonomous mobile robot runs around a pedestrian, it is necessary to avoid the pedestrian safely. For that purpose, it is important to predict the trajectory of the pedestrian and take the avoidance action an earlier stage. In this method, the LSTM encoder-decoder model is used to predict the complex behavior of pedestrians. The input of the model is the obstacle information around pedestrian obtained by range-finder sensor and the past position of the pedestrian. Through experiments in a simulation environment, we show that the proposed model can predict the behavior of pedestrians affected by surrounding obstacles.